Meta pauses AI programme that tracked employee mouse clicks after internal data Leak
Meta suspended its internal AI training program after it was found that sensitive employee data, including private conversations and performance records, was accessible to all staff at the company.
What Happened: Meta’s Internal AI Data Made Available to All Employees
Meta is pausing an internal AI training program after a security incident exposed sensitive employee data across the company, Business Insider first reported, and Reuters later confirmed the details.
The program, called the Model Talent Initiative (MCI), was launched in April to improve Meta’s AI models by collecting keystrokes, mouse movements and clicks on the work computers of US-based employees.
Data exposed in the incident included full prompts and transcripts, private conversations, contacts and performance data, and DSS sensitivity ratings on a scale of one to four.
The incident was classified as SEV 2 on Meta’s internal severity scale of zero to five; zero was the most serious.
Meta Confirms Investigation, No Evidence of Abuse
A Meta spokesperson confirmed the incident Business Content and said the company was investigating the matter.
“We have carefully designed this program with privacy safeguards, and while we currently have no indication that any data has been improperly accessed by Meta employees, we are pausing the program while we investigate,” the spokesperson told BI.
Another company spokeswoman, Tracy Clayton, echoed the same statement. Reuterswhile declining to specify how long the program will be suspended.
Despite the announced pause, a source said Reuters The vehicle was still actively recording data as of Monday afternoon. The company acknowledged that halting the program for all employees would take time to implement.
Employees’ Reaction: “I’m Angry”
Screenshots of internal discussions obtained by Business Content Show employees responding with alarm.
“I am outraged,” one Meta employee wrote in an internal group on Monday. “I don’t see any evidence of malicious access, but it is extremely frustrating that this data was not locked as originally promised,” the same employee continued.
Another Meta employee, writing in SEV’s internal discussion thread, raised broader concerns about the scope of what was seized.
The employee requested further investigation of the issues, saying, “Like thousands of employees, I had access to both personal tax information and medical information through my work computer. We were told that this data would be protected and used only for valid business purposes after aggressive filtering.” he wrote.
The MCI program was already controversial before the leak. When it was introduced as mandatory for most staff in April, employees expressed dismay at having their digital activities recorded for AI training purposes.
Previous Reuters Investigation Raises Red Flags
The latest incident is as follows: Reuters investigation In May, it was revealed that the program at Meta was collecting more information than initially disclosed to employees and that the data was being stored in an unencrypted form.
These findings had already raised privacy concerns internally before the current leak.
Part of a Broader Pattern of Security Incidents in Meta
The data exposure is the latest in a series of security-related incidents at Meta. Last month, a flaw was found in Meta’s AI chatbot that allowed users to take over multiple Instagram accounts.
In March, a rogue AI agent caused a serious internal incident, The Information reports.
What is Meta’s Model Talent Initiative?
The Model Talent Initiative was introduced in April as a mandatory program for most US-based Meta employees. The tool is designed to capture mouse movements, clicks, and keystrokes on employees’ computers to create training data for Meta’s AI models.
The current pause was triggered after an employee filed a formal SEV, or high-priority security incident report, over employee data exposure.
Reporting is based on documents reviewed by Reuters and screenshots obtained by Business Insider.
